Data your team can defend in a meeting.

What “trusted data” actually means here

“Trusted” is an overused word in analytics. In Sluice it has a precise, testable definition — four guarantees that hold for every number on your dashboard:

• The AI plans, it never writes SQL. The model produces a structured plan — metrics, dimensions, filters, joins — over your real schema. A deterministic compiler turns that plan into SQL, so the same plan always compiles to the same query.
• A human approves the plan. Nothing runs until you review and approve it. The plan is readable, versioned, and auditable — not a prompt buried in a black box.
• Every number is provable to source. Each KPI can be traced back to the raw rows it aggregates. The total on the chart equals the sum of the source — you can check it.
• Governed by default. PII is masked, access is role-scoped, and the whole pipeline is reproducible from raw data any time.

Why AI-written SQL isn’t trustworthy

When an AI writes the query directly, two identical questions can produce two different queries — and neither is reproducible or auditable. That’s fine for casual exploration, but it’s not something you can put in a board deck, a compliance report, or a customer invoice. Sluice keeps the AI on the planning side of the line and lets deterministic code own the execution side. You get the speed of AI with the accountability of engineered pipelines.

How a trusted answer is built

Where this matters most

Trusted data is not a nice-to-have when the number carries money or risk: freight and GST audits, financial KPIs, regulated reporting, and any figure a customer or auditor might challenge. If you can’t trace it to source, you can’t defend it. Sluice is built so you always can.